Fig. 4: Adaptation to anticipated future needs.
From: Computational models of episodic-like memory in food-caching birds

A The preference of caching worms in a given tray depends on the red caching weights \({w}_{fx}^{{{{{{{{\rm{cache}}}}}}}}}\). B One day later, when the bird inspects the caching tray, the memory of the cached food is reactivated and, in particular, the pre- and postsynaptic neurons of the caching weights. After discovering that the cached food item has disappeared, a feedback signal acts as a third factor in a neoHebbian plasticity rule, thereby decreasing the caching weight. C The Planning-By-Replay Model memorizes the sequence of events (indicated with film strip). The caching preference is determined by searching for positions in memory that match best the current context (indicated with black triangle) and evaluating the preference of a caching action under the assumption that the sequence of future events resembles the sequence of events following the matched positions. All other components of the Planning-By-Replay Model are the same as in the Plastic Caching Model.